Save the date! Google I/O returns May 18-20 Register now

Transformer model for language understanding

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

This tutorial trains a Transformer model to translate a Portuguese to English dataset. This is an advanced example that assumes knowledge of text generation and attention.

The core idea behind the Transformer model is self-attention—the ability to attend to different positions of the input sequence to compute a representation of that sequence. Transformer creates stacks of self-attention layers and is explained below in the sections Scaled dot product attention and Multi-head attention.

A transformer model handles variable-sized input using stacks of self-attention layers instead of RNNs or CNNs. This general architecture has a number of advantages:

  • It makes no assumptions about the temporal/spatial relationships across the data. This is ideal for processing a set of objects (for example, StarCraft units).
  • Layer outputs can be calculated in parallel, instead of a series like an RNN.
  • Distant items can affect each other's output without passing through many RNN-steps, or convolution layers (see Scene Memory Transformer for example).
  • It can learn long-range dependencies. This is a challenge in many sequence tasks.

The downsides of this architecture are:

  • For a time-series, the output for a time-step is calculated from the entire history instead of only the inputs and current hidden-state. This may be less efficient.
  • If the input does have a temporal/spatial relationship, like text, some positional encoding must be added or the model will effectively see a bag of words.

After training the model in this notebook, you will be able to input a Portuguese sentence and return the English translation.

Attention heatmap

Setup

pip install -q tensorflow_datasets
pip install -q tensorflow_text

import collections
import logging
import os
import pathlib
import re
import string
import sys
import time

import numpy as np
import matplotlib.pyplot as plt

import tensorflow_datasets as tfds
import tensorflow_text as text
import tensorflow as tf

logging.getLogger('tensorflow').setLevel(logging.ERROR)  # suppress warnings

Download the Dataset

Use TensorFlow datasets to load the Portuguese-English translation dataset from the TED Talks Open Translation Project.

This dataset contains approximately 50000 training examples, 1100 validation examples, and 2000 test examples.

examples, metadata = tfds.load('ted_hrlr_translate/pt_to_en', with_info=True,
                               as_supervised=True)
train_examples, val_examples = examples['train'], examples['validation']

The tf.data.Dataset object returned by TensorFlow datasets yields pairs of text examples:

for pt_examples, en_examples in train_examples.batch(3).take(1):
  for pt in pt_examples.numpy():
    print(pt.decode('utf-8'))

  print()

  for en in en_examples.numpy():
    print(en.decode('utf-8'))

e quando melhoramos a procura , tiramos a única vantagem da impressão , que é a serendipidade .
mas e se estes fatores fossem ativos ?
mas eles não tinham a curiosidade de me testar .

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .

Text tokenization & detokenization

You can't train a model directly on text. The text needs to be converted to some numeric representation first. Typically, you convert the text to sequences of token IDs, which are as indexes into an embedding.

One popular implementation is demonstrated in the Subword tokenizer tutorial builds subword tokenizers (text.BertTokenizer) optimized for this dataset and exports them in a saved_model.

Download and unzip and import the saved_model:

model_name = "ted_hrlr_translate_pt_en_converter"
tf.keras.utils.get_file(
    f"{model_name}.zip",
    f"https://storage.googleapis.com/download.tensorflow.org/models/{model_name}.zip",
    cache_dir='.', cache_subdir='', extract=True
)

Downloading data from https://storage.googleapis.com/download.tensorflow.org/models/ted_hrlr_translate_pt_en_converter.zip
188416/184801 [==============================] - 0s 0us/step
'./ted_hrlr_translate_pt_en_converter.zip'

tokenizers = tf.saved_model.load(model_name)

The tf.saved_model contains two text tokenizers, one for English and one for Portugese. Both have the same methods:

[item for item in dir(tokenizers.en) if not item.startswith('_')]

['detokenize',
 'get_reserved_tokens',
 'get_vocab_path',
 'get_vocab_size',
 'lookup',
 'tokenize',
 'tokenizer',
 'vocab']

The tokenize method converts a batch of strings to a padded-batch of token IDs. This method splits punctuation, lowercases and unicode-normalizes the input before tokenizing. That standardization is not visible here because the input data is already standardized.

for en in en_examples.numpy():
  print(en.decode('utf-8'))

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .

encoded = tokenizers.en.tokenize(en_examples)

for row in encoded.to_list():
  print(row)

[2, 72, 117, 79, 1259, 1491, 2362, 13, 79, 150, 184, 311, 71, 103, 2308, 74, 2679, 13, 148, 80, 55, 4840, 1434, 2423, 540, 15, 3]
[2, 87, 90, 107, 76, 129, 1852, 30, 3]
[2, 87, 83, 149, 50, 9, 56, 664, 85, 2512, 15, 3]

The detokenize method attempts to convert these token IDs back to human readable text: 

round_trip = tokenizers.en.detokenize(encoded)
for line in round_trip.numpy():
  print(line.decode('utf-8'))

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n ' t test for curiosity .

The lower level lookup method converts from token-IDs to token text:

tokens = tokenizers.en.lookup(encoded)
tokens

<tf.RaggedTensor [[b'[START]', b'and', b'when', b'you', b'improve', b'search', b'##ability', b',', b'you', b'actually', b'take', b'away', b'the', b'one', b'advantage', b'of', b'print', b',', b'which', b'is', b's', b'##ere', b'##nd', b'##ip', b'##ity', b'.', b'[END]'], [b'[START]', b'but', b'what', b'if', b'it', b'were', b'active', b'?', b'[END]'], [b'[START]', b'but', b'they', b'did', b'n', b"'", b't', b'test', b'for', b'curiosity', b'.', b'[END]']]>

Here you can see the "subword" aspect of the tokenizers. The word "searchability" is decomposed into "search ##ability" and the word "serindipity" into "s ##ere ##nd ##ip ##ity"

Setup input pipeline

To build an input pipeline suitable for training you'll apply some transformations to the dataset.

This function will be used to encode the batches of raw text:

def tokenize_pairs(pt, en):
    pt = tokenizers.pt.tokenize(pt)
    # Convert from ragged to dense, padding with zeros.
    pt = pt.to_tensor()

    en = tokenizers.en.tokenize(en)
    # Convert from ragged to dense, padding with zeros.
    en = en.to_tensor()
    return pt, en

Here's a simple input pipeline that processes, shuffles and batches the data:

BUFFER_SIZE = 20000
BATCH_SIZE = 64

def make_batches(ds):
  return (
      ds
      .cache()
      .shuffle(BUFFER_SIZE)
      .batch(BATCH_SIZE)
      .map(tokenize_pairs, num_parallel_calls=tf.data.AUTOTUNE)
      .prefetch(tf.data.AUTOTUNE))


train_batches = make_batches(train_examples)
val_batches = make_batches(val_examples)

Positional encoding

Since this model doesn't contain any recurrence or convolution, positional encoding is added to give the model some information about the relative position of the words in the sentence.

The positional encoding vector is added to the embedding vector. Embeddings represent a token in a d-dimensional space where tokens with similar meaning will be closer to each other. But the embeddings do not encode the relative position of words in a sentence. So after adding the positional encoding, words will be closer to each other based on the similarity of their meaning and their position in the sentence, in the d-dimensional space.

The formula for calculating the positional encoding is as follows:

$$\Large{PE_{(pos, 2i)} = \sin(pos / 10000^{2i / d_{model} })} $$
$$\Large{PE_{(pos, 2i+1)} = \cos(pos / 10000^{2i / d_{model} })} $$ 
def get_angles(pos, i, d_model):
  angle_rates = 1 / np.power(10000, (2 * (i//2)) / np.float32(d_model))
  return pos * angle_rates

def positional_encoding(position, d_model):
  angle_rads = get_angles(np.arange(position)[:, np.newaxis],
                          np.arange(d_model)[np.newaxis, :],
                          d_model)

  # apply sin to even indices in the array; 2i
  angle_rads[:, 0::2] = np.sin(angle_rads[:, 0::2])

  # apply cos to odd indices in the array; 2i+1
  angle_rads[:, 1::2] = np.cos(angle_rads[:, 1::2])

  pos_encoding = angle_rads[np.newaxis, ...]

  return tf.cast(pos_encoding, dtype=tf.float32)

n, d = 2048, 512
pos_encoding = positional_encoding(n, d)
print(pos_encoding.shape)
pos_encoding = pos_encoding[0]

# Juggle the dimensions for the plot
pos_encoding = tf.reshape(pos_encoding, (n, d//2, 2))
pos_encoding = tf.transpose(pos_encoding, (2, 1, 0))
pos_encoding = tf.reshape(pos_encoding, (d, n))

plt.pcolormesh(pos_encoding, cmap='RdBu')
plt.ylabel('Depth')
plt.xlabel('Position')
plt.colorbar()
plt.show()

(1, 2048, 512)

png

Masking

Mask all the pad tokens in the batch of sequence. It ensures that the model does not treat padding as the input. The mask indicates where pad value 0 is present: it outputs a 1 at those locations, and a 0 otherwise.

def create_padding_mask(seq):
  seq = tf.cast(tf.math.equal(seq, 0), tf.float32)

  # add extra dimensions to add the padding
  # to the attention logits.
  return seq[:, tf.newaxis, tf.newaxis, :]  # (batch_size, 1, 1, seq_len)

x = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]])
create_padding_mask(x)

<tf.Tensor: shape=(3, 1, 1, 5), dtype=float32, numpy=
array([[[[0., 0., 1., 1., 0.]]],


       [[[0., 0., 0., 1., 1.]]],


       [[[1., 1., 1., 0., 0.]]]], dtype=float32)>

The look-ahead mask is used to mask the future tokens in a sequence. In other words, the mask indicates which entries should not be used.

This means that to predict the third word, only the first and second word will be used. Similarly to predict the fourth word, only the first, second and the third word will be used and so on.

def create_look_ahead_mask(size):
  mask = 1 - tf.linalg.band_part(tf.ones((size, size)), -1, 0)
  return mask  # (seq_len, seq_len)

x = tf.random.uniform((1, 3))
temp = create_look_ahead_mask(x.shape[1])
temp

<tf.Tensor: shape=(3, 3), dtype=float32, numpy=
array([[0., 1., 1.],
       [0., 0., 1.],
       [0., 0., 0.]], dtype=float32)>

Scaled dot product attention

scaled_dot_product_attention

The attention function used by the transformer takes three inputs: Q (query), K (key), V (value). The equation used to calculate the attention weights is:

$$\Large{Attention(Q, K, V) = softmax_k\left(\frac{QK^T}{\sqrt{d_k} }\right) V} $$

The dot-product attention is scaled by a factor of square root of the depth. This is done because for large values of depth, the dot product grows large in magnitude pushing the softmax function where it has small gradients resulting in a very hard softmax.

For example, consider that Q and K have a mean of 0 and variance of 1. Their matrix multiplication will have a mean of 0 and variance of dk. So the square root of dk is used for scaling, so you get a consistent variance regardless of the value of dk. If the variance is too low the output may be too flat to optimize effectively. If the variance is too high the softmax may saturate at initilization making it dificult to learn.

The mask is multiplied with -1e9 (close to negative infinity). This is done because the mask is summed with the scaled matrix multiplication of Q and K and is applied immediately before a softmax. The goal is to zero out these cells, and large negative inputs to softmax are near zero in the output.

def scaled_dot_product_attention(q, k, v, mask):
  """Calculate the attention weights.
  q, k, v must have matching leading dimensions.
  k, v must have matching penultimate dimension, i.e.: seq_len_k = seq_len_v.
  The mask has different shapes depending on its type(padding or look ahead)
  but it must be broadcastable for addition.

  Args:
    q: query shape == (..., seq_len_q, depth)
    k: key shape == (..., seq_len_k, depth)
    v: value shape == (..., seq_len_v, depth_v)
    mask: Float tensor with shape broadcastable
          to (..., seq_len_q, seq_len_k). Defaults to None.

  Returns:
    output, attention_weights
  """

  matmul_qk = tf.matmul(q, k, transpose_b=True)  # (..., seq_len_q, seq_len_k)

  # scale matmul_qk
  dk = tf.cast(tf.shape(k)[-1], tf.float32)
  scaled_attention_logits = matmul_qk / tf.math.sqrt(dk)

  # add the mask to the scaled tensor.
  if mask is not None:
    scaled_attention_logits += (mask * -1e9)

  # softmax is normalized on the last axis (seq_len_k) so that the scores
  # add up to 1.
  attention_weights = tf.nn.softmax(scaled_attention_logits, axis=-1)  # (..., seq_len_q, seq_len_k)

  output = tf.matmul(attention_weights, v)  # (..., seq_len_q, depth_v)

  return output, attention_weights

As the softmax normalization is done on K, its values decide the amount of importance given to Q.

The output represents the multiplication of the attention weights and the V (value) vector. This ensures that the words you want to focus on are kept as-is and the irrelevant words are flushed out.

def print_out(q, k, v):
  temp_out, temp_attn = scaled_dot_product_attention(
      q, k, v, None)
  print('Attention weights are:')
  print(temp_attn)
  print('Output is:')
  print(temp_out)

np.set_printoptions(suppress=True)

temp_k = tf.constant([[10, 0, 0],
                      [0, 10, 0],
                      [0, 0, 10],
                      [0, 0, 10]], dtype=tf.float32)  # (4, 3)

temp_v = tf.constant([[1, 0],
                      [10, 0],
                      [100, 5],
                      [1000, 6]], dtype=tf.float32)  # (4, 2)

# This `query` aligns with the second `key`,
# so the second `value` is returned.
temp_q = tf.constant([[0, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0. 1. 0. 0.]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[10.  0.]], shape=(1, 2), dtype=float32)

# This query aligns with a repeated key (third and fourth),
# so all associated values get averaged.
temp_q = tf.constant([[0, 0, 10]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0.  0.  0.5 0.5]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[550.    5.5]], shape=(1, 2), dtype=float32)

# This query aligns equally with the first and second key,
# so their values get averaged.
temp_q = tf.constant([[10, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0.5 0.5 0.  0. ]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[5.5 0. ]], shape=(1, 2), dtype=float32)

Pass all the queries together.

temp_q = tf.constant([[0, 0, 10],
                      [0, 10, 0],
                      [10, 10, 0]], dtype=tf.float32)  # (3, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor(
[[0.  0.  0.5 0.5]
 [0.  1.  0.  0. ]
 [0.5 0.5 0.  0. ]], shape=(3, 4), dtype=float32)
Output is:
tf.Tensor(
[[550.    5.5]
 [ 10.    0. ]
 [  5.5   0. ]], shape=(3, 2), dtype=float32)

Multi-head attention

multi-head attention

Multi-head attention consists of four parts:

  • Linear layers and split into heads.
  • Scaled dot-product attention.
  • Concatenation of heads.
  • Final linear layer.

Each multi-head attention block gets three inputs; Q (query), K (key), V (value). These are put through linear (Dense) layers and split up into multiple heads.

The scaled_dot_product_attention defined above is applied to each head (broadcasted for efficiency). An appropriate mask must be used in the attention step. The attention output for each head is then concatenated (using tf.transpose, and tf.reshape) and put through a final Dense layer.

Instead of one single attention head, Q, K, and V are split into multiple heads because it allows the model to jointly attend to information at different positions from different representational spaces. After the split each head has a reduced dimensionality, so the total computation cost is the same as a single head attention with full dimensionality.

class MultiHeadAttention(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads):
    super(MultiHeadAttention, self).__init__()
    self.num_heads = num_heads
    self.d_model = d_model

    assert d_model % self.num_heads == 0

    self.depth = d_model // self.num_heads

    self.wq = tf.keras.layers.Dense(d_model)
    self.wk = tf.keras.layers.Dense(d_model)
    self.wv = tf.keras.layers.Dense(d_model)

    self.dense = tf.keras.layers.Dense(d_model)

  def split_heads(self, x, batch_size):
    """Split the last dimension into (num_heads, depth).
    Transpose the result such that the shape is (batch_size, num_heads, seq_len, depth)
    """
    x = tf.reshape(x, (batch_size, -1, self.num_heads, self.depth))
    return tf.transpose(x, perm=[0, 2, 1, 3])

  def call(self, v, k, q, mask):
    batch_size = tf.shape(q)[0]

    q = self.wq(q)  # (batch_size, seq_len, d_model)
    k = self.wk(k)  # (batch_size, seq_len, d_model)
    v = self.wv(v)  # (batch_size, seq_len, d_model)

    q = self.split_heads(q, batch_size)  # (batch_size, num_heads, seq_len_q, depth)
    k = self.split_heads(k, batch_size)  # (batch_size, num_heads, seq_len_k, depth)
    v = self.split_heads(v, batch_size)  # (batch_size, num_heads, seq_len_v, depth)

    # scaled_attention.shape == (batch_size, num_heads, seq_len_q, depth)
    # attention_weights.shape == (batch_size, num_heads, seq_len_q, seq_len_k)
    scaled_attention, attention_weights = scaled_dot_product_attention(
        q, k, v, mask)

    scaled_attention = tf.transpose(scaled_attention, perm=[0, 2, 1, 3])  # (batch_size, seq_len_q, num_heads, depth)

    concat_attention = tf.reshape(scaled_attention,
                                  (batch_size, -1, self.d_model))  # (batch_size, seq_len_q, d_model)

    output = self.dense(concat_attention)  # (batch_size, seq_len_q, d_model)

    return output, attention_weights

Create a MultiHeadAttention layer to try out. At each location in the sequence, y, the MultiHeadAttention runs all 8 attention heads across all other locations in the sequence, returning a new vector of the same length at each location.

temp_mha = MultiHeadAttention(d_model=512, num_heads=8)
y = tf.random.uniform((1, 60, 512))  # (batch_size, encoder_sequence, d_model)
out, attn = temp_mha(y, k=y, q=y, mask=None)
out.shape, attn.shape

(TensorShape([1, 60, 512]), TensorShape([1, 8, 60, 60]))

Point wise feed forward network

Point wise feed forward network consists of two fully-connected layers with a ReLU activation in between.

def point_wise_feed_forward_network(d_model, dff):
  return tf.keras.Sequential([
      tf.keras.layers.Dense(dff, activation='relu'),  # (batch_size, seq_len, dff)
      tf.keras.layers.Dense(d_model)  # (batch_size, seq_len, d_model)
  ])

sample_ffn = point_wise_feed_forward_network(512, 2048)
sample_ffn(tf.random.uniform((64, 50, 512))).shape

TensorShape([64, 50, 512])

Encoder and decoder

transformer

The transformer model follows the same general pattern as a standard sequence to sequence with attention model.

  • The input sentence is passed through N encoder layers that generates an output for each word/token in the sequence.
  • The decoder attends on the encoder's output and its own input (self-attention) to predict the next word.

Encoder layer

Each encoder layer consists of sublayers:

  1. Multi-head attention (with padding mask)
  2. Point wise feed forward networks.

Each of these sublayers has a residual connection around it followed by a layer normalization. Residual connections help in avoiding the vanishing gradient problem in deep networks.

The output of each sublayer is LayerNorm(x + Sublayer(x)). The normalization is done on the d_model (last) axis. There are N encoder layers in the transformer.

class EncoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(EncoderLayer, self).__init__()

    self.mha = MultiHeadAttention(d_model, num_heads)
    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    attn_output, _ = self.mha(x, x, x, mask)  # (batch_size, input_seq_len, d_model)
    attn_output = self.dropout1(attn_output, training=training)
    out1 = self.layernorm1(x + attn_output)  # (batch_size, input_seq_len, d_model)

    ffn_output = self.ffn(out1)  # (batch_size, input_seq_len, d_model)
    ffn_output = self.dropout2(ffn_output, training=training)
    out2 = self.layernorm2(out1 + ffn_output)  # (batch_size, input_seq_len, d_model)

    return out2

sample_encoder_layer = EncoderLayer(512, 8, 2048)

sample_encoder_layer_output = sample_encoder_layer(
    tf.random.uniform((64, 43, 512)), False, None)

sample_encoder_layer_output.shape  # (batch_size, input_seq_len, d_model)

TensorShape([64, 43, 512])

Decoder layer

Each decoder layer consists of sublayers:

  1. Masked multi-head attention (with look ahead mask and padding mask)
  2. Multi-head attention (with padding mask). V (value) and K (key) receive the encoder output as inputs. Q (query) receives the output from the masked multi-head attention sublayer.
  3. Point wise feed forward networks

Each of these sublayers has a residual connection around it followed by a layer normalization. The output of each sublayer is LayerNorm(x + Sublayer(x)). The normalization is done on the d_model (last) axis.

There are N decoder layers in the transformer.

As Q receives the output from decoder's first attention block, and K receives the encoder output, the attention weights represent the importance given to the decoder's input based on the encoder's output. In other words, the decoder predicts the next word by looking at the encoder output and self-attending to its own output. See the demonstration above in the scaled dot product attention section.

class DecoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(DecoderLayer, self).__init__()

    self.mha1 = MultiHeadAttention(d_model, num_heads)
    self.mha2 = MultiHeadAttention(d_model, num_heads)

    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm3 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)
    self.dropout3 = tf.keras.layers.Dropout(rate)

  def call(self, x, enc_output, training,
           look_ahead_mask, padding_mask):
    # enc_output.shape == (batch_size, input_seq_len, d_model)

    attn1, attn_weights_block1 = self.mha1(x, x, x, look_ahead_mask)  # (batch_size, target_seq_len, d_model)
    attn1 = self.dropout1(attn1, training=training)
    out1 = self.layernorm1(attn1 + x)

    attn2, attn_weights_block2 = self.mha2(
        enc_output, enc_output, out1, padding_mask)  # (batch_size, target_seq_len, d_model)
    attn2 = self.dropout2(attn2, training=training)
    out2 = self.layernorm2(attn2 + out1)  # (batch_size, target_seq_len, d_model)

    ffn_output = self.ffn(out2)  # (batch_size, target_seq_len, d_model)
    ffn_output = self.dropout3(ffn_output, training=training)
    out3 = self.layernorm3(ffn_output + out2)  # (batch_size, target_seq_len, d_model)

    return out3, attn_weights_block1, attn_weights_block2

sample_decoder_layer = DecoderLayer(512, 8, 2048)

sample_decoder_layer_output, _, _ = sample_decoder_layer(
    tf.random.uniform((64, 50, 512)), sample_encoder_layer_output,
    False, None, None)

sample_decoder_layer_output.shape  # (batch_size, target_seq_len, d_model)

TensorShape([64, 50, 512])

Encoder

The Encoder consists of:

  1. Input Embedding
  2. Positional Encoding
  3. N encoder layers

The input is put through an embedding which is summed with the positional encoding. The output of this summation is the input to the encoder layers. The output of the encoder is the input to the decoder.

class Encoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Encoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(input_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding,
                                            self.d_model)

    self.enc_layers = [EncoderLayer(d_model, num_heads, dff, rate)
                       for _ in range(num_layers)]

    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    seq_len = tf.shape(x)[1]

    # adding embedding and position encoding.
    x = self.embedding(x)  # (batch_size, input_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x = self.enc_layers[i](x, training, mask)

    return x  # (batch_size, input_seq_len, d_model)

sample_encoder = Encoder(num_layers=2, d_model=512, num_heads=8,
                         dff=2048, input_vocab_size=8500,
                         maximum_position_encoding=10000)
temp_input = tf.random.uniform((64, 62), dtype=tf.int64, minval=0, maxval=200)

sample_encoder_output = sample_encoder(temp_input, training=False, mask=None)

print(sample_encoder_output.shape)  # (batch_size, input_seq_len, d_model)

(64, 62, 512)

Decoder

The Decoder consists of:

  1. Output Embedding
  2. Positional Encoding
  3. N decoder layers

The target is put through an embedding which is summed with the positional encoding. The output of this summation is the input to the decoder layers. The output of the decoder is the input to the final linear layer.

class Decoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, target_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Decoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(target_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding, d_model)

    self.dec_layers = [DecoderLayer(d_model, num_heads, dff, rate)
                       for _ in range(num_layers)]
    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, enc_output, training,
           look_ahead_mask, padding_mask):

    seq_len = tf.shape(x)[1]
    attention_weights = {}

    x = self.embedding(x)  # (batch_size, target_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x, block1, block2 = self.dec_layers[i](x, enc_output, training,
                                             look_ahead_mask, padding_mask)

      attention_weights[f'decoder_layer{i+1}_block1'] = block1
      attention_weights[f'decoder_layer{i+1}_block2'] = block2

    # x.shape == (batch_size, target_seq_len, d_model)
    return x, attention_weights

sample_decoder = Decoder(num_layers=2, d_model=512, num_heads=8,
                         dff=2048, target_vocab_size=8000,
                         maximum_position_encoding=5000)
temp_input = tf.random.uniform((64, 26), dtype=tf.int64, minval=0, maxval=200)

output, attn = sample_decoder(temp_input,
                              enc_output=sample_encoder_output,
                              training=False,
                              look_ahead_mask=None,
                              padding_mask=None)

output.shape, attn['decoder_layer2_block2'].shape

(TensorShape([64, 26, 512]), TensorShape([64, 8, 26, 62]))

Create the Transformer

Transformer consists of the encoder, decoder and a final linear layer. The output of the decoder is the input to the linear layer and its output is returned.

class Transformer(tf.keras.Model):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size,
               target_vocab_size, pe_input, pe_target, rate=0.1):
    super(Transformer, self).__init__()

    self.tokenizer = Encoder(num_layers, d_model, num_heads, dff,
                             input_vocab_size, pe_input, rate)

    self.decoder = Decoder(num_layers, d_model, num_heads, dff,
                           target_vocab_size, pe_target, rate)

    self.final_layer = tf.keras.layers.Dense(target_vocab_size)

  def call(self, inp, tar, training, enc_padding_mask,
           look_ahead_mask, dec_padding_mask):

    enc_output = self.tokenizer(inp, training, enc_padding_mask)  # (batch_size, inp_seq_len, d_model)

    # dec_output.shape == (batch_size, tar_seq_len, d_model)
    dec_output, attention_weights = self.decoder(
        tar, enc_output, training, look_ahead_mask, dec_padding_mask)

    final_output = self.final_layer(dec_output)  # (batch_size, tar_seq_len, target_vocab_size)

    return final_output, attention_weights

sample_transformer = Transformer(
    num_layers=2, d_model=512, num_heads=8, dff=2048,
    input_vocab_size=8500, target_vocab_size=8000,
    pe_input=10000, pe_target=6000)

temp_input = tf.random.uniform((64, 38), dtype=tf.int64, minval=0, maxval=200)
temp_target = tf.random.uniform((64, 36), dtype=tf.int64, minval=0, maxval=200)

fn_out, _ = sample_transformer(temp_input, temp_target, training=False,
                               enc_padding_mask=None,
                               look_ahead_mask=None,
                               dec_padding_mask=None)

fn_out.shape  # (batch_size, tar_seq_len, target_vocab_size)

TensorShape([64, 36, 8000])

Set hyperparameters

To keep this example small and relatively fast, the values for num_layers, d_model, and dff have been reduced.

The values used in the base model of transformer were; num_layers=6, d_model = 512, dff = 2048. See the paper for all the other versions of the transformer.

num_layers = 4
d_model = 128
dff = 512
num_heads = 8
dropout_rate = 0.1

Optimizer

Use the Adam optimizer with a custom learning rate scheduler according to the formula in the paper.

$$\Large{lrate = d_{model}^{-0.5} * \min(step{\_}num^{-0.5}, step{\_}num \cdot warmup{\_}steps^{-1.5})}$$ 
class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):
  def __init__(self, d_model, warmup_steps=4000):
    super(CustomSchedule, self).__init__()

    self.d_model = d_model
    self.d_model = tf.cast(self.d_model, tf.float32)

    self.warmup_steps = warmup_steps

  def __call__(self, step):
    arg1 = tf.math.rsqrt(step)
    arg2 = step * (self.warmup_steps ** -1.5)

    return tf.math.rsqrt(self.d_model) * tf.math.minimum(arg1, arg2)

learning_rate = CustomSchedule(d_model)

optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.9, beta_2=0.98,
                                     epsilon=1e-9)

temp_learning_rate_schedule = CustomSchedule(d_model)

plt.plot(temp_learning_rate_schedule(tf.range(40000, dtype=tf.float32)))
plt.ylabel("Learning Rate")
plt.xlabel("Train Step")

Text(0.5, 0, 'Train Step')

png

Loss and metrics

Since the target sequences are padded, it is important to apply a padding mask when calculating the loss.

loss_object = tf.keras.losses.SparseCategoricalCrossentropy(
    from_logits=True, reduction='none')

def loss_function(real, pred):
  mask = tf.math.logical_not(tf.math.equal(real, 0))
  loss_ = loss_object(real, pred)

  mask = tf.cast(mask, dtype=loss_.dtype)
  loss_ *= mask

  return tf.reduce_sum(loss_)/tf.reduce_sum(mask)


def accuracy_function(real, pred):
  accuracies = tf.equal(real, tf.argmax(pred, axis=2))

  mask = tf.math.logical_not(tf.math.equal(real, 0))
  accuracies = tf.math.logical_and(mask, accuracies)

  accuracies = tf.cast(accuracies, dtype=tf.float32)
  mask = tf.cast(mask, dtype=tf.float32)
  return tf.reduce_sum(accuracies)/tf.reduce_sum(mask)

train_loss = tf.keras.metrics.Mean(name='train_loss')
train_accuracy = tf.keras.metrics.Mean(name='train_accuracy')

Training and checkpointing

transformer = Transformer(
    num_layers=num_layers,
    d_model=d_model,
    num_heads=num_heads,
    dff=dff,
    input_vocab_size=tokenizers.pt.get_vocab_size(),
    target_vocab_size=tokenizers.en.get_vocab_size(),
    pe_input=1000,
    pe_target=1000,
    rate=dropout_rate)

def create_masks(inp, tar):
  # Encoder padding mask
  enc_padding_mask = create_padding_mask(inp)

  # Used in the 2nd attention block in the decoder.
  # This padding mask is used to mask the encoder outputs.
  dec_padding_mask = create_padding_mask(inp)

  # Used in the 1st attention block in the decoder.
  # It is used to pad and mask future tokens in the input received by
  # the decoder.
  look_ahead_mask = create_look_ahead_mask(tf.shape(tar)[1])
  dec_target_padding_mask = create_padding_mask(tar)
  combined_mask = tf.maximum(dec_target_padding_mask, look_ahead_mask)

  return enc_padding_mask, combined_mask, dec_padding_mask

Create the checkpoint path and the checkpoint manager. This will be used to save checkpoints every n epochs.

checkpoint_path = "./checkpoints/train"

ckpt = tf.train.Checkpoint(transformer=transformer,
                           optimizer=optimizer)

ckpt_manager = tf.train.CheckpointManager(ckpt, checkpoint_path, max_to_keep=5)

# if a checkpoint exists, restore the latest checkpoint.
if ckpt_manager.latest_checkpoint:
  ckpt.restore(ckpt_manager.latest_checkpoint)
  print('Latest checkpoint restored!!')

The target is divided into tar_inp and tar_real. tar_inp is passed as an input to the decoder. tar_real is that same input shifted by 1: At each location in tar_input, tar_real contains the next token that should be predicted.

For example, sentence = "SOS A lion in the jungle is sleeping EOS"

tar_inp = "SOS A lion in the jungle is sleeping"

tar_real = "A lion in the jungle is sleeping EOS"

The transformer is an auto-regressive model: it makes predictions one part at a time, and uses its output so far to decide what to do next.

During training this example uses teacher-forcing (like in the text generation tutorial). Teacher forcing is passing the true output to the next time step regardless of what the model predicts at the current time step.

As the transformer predicts each word, self-attention allows it to look at the previous words in the input sequence to better predict the next word.

To prevent the model from peeking at the expected output the model uses a look-ahead mask.

EPOCHS = 20

# The @tf.function trace-compiles train_step into a TF graph for faster
# execution. The function specializes to the precise shape of the argument
# tensors. To avoid re-tracing due to the variable sequence lengths or variable
# batch sizes (the last batch is smaller), use input_signature to specify
# more generic shapes.

train_step_signature = [
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
]


@tf.function(input_signature=train_step_signature)
def train_step(inp, tar):
  tar_inp = tar[:, :-1]
  tar_real = tar[:, 1:]

  enc_padding_mask, combined_mask, dec_padding_mask = create_masks(inp, tar_inp)

  with tf.GradientTape() as tape:
    predictions, _ = transformer(inp, tar_inp,
                                 True,
                                 enc_padding_mask,
                                 combined_mask,
                                 dec_padding_mask)
    loss = loss_function(tar_real, predictions)

  gradients = tape.gradient(loss, transformer.trainable_variables)
  optimizer.apply_gradients(zip(gradients, transformer.trainable_variables))

  train_loss(loss)
  train_accuracy(accuracy_function(tar_real, predictions))

Portuguese is used as the input language and English is the target language.

for epoch in range(EPOCHS):
  start = time.time()

  train_loss.reset_states()
  train_accuracy.reset_states()

  # inp -> portuguese, tar -> english
  for (batch, (inp, tar)) in enumerate(train_batches):
    train_step(inp, tar)

    if batch % 50 == 0:
      print(f'Epoch {epoch + 1} Batch {batch} Loss {train_loss.result():.4f} Accuracy {train_accuracy.result():.4f}')

  if (epoch + 1) % 5 == 0:
    ckpt_save_path = ckpt_manager.save()
    print(f'Saving checkpoint for epoch {epoch+1} at {ckpt_save_path}')

  print(f'Epoch {epoch + 1} Loss {train_loss.result():.4f} Accuracy {train_accuracy.result():.4f}')

  print(f'Time taken for 1 epoch: {time.time() - start:.2f} secs\n')

Epoch 1 Batch 0 Loss 8.8704 Accuracy 0.0081
Epoch 1 Batch 50 Loss 8.8161 Accuracy 0.0091
Epoch 1 Batch 100 Loss 8.7176 Accuracy 0.0225
Epoch 1 Batch 150 Loss 8.6033 Accuracy 0.0354
Epoch 1 Batch 200 Loss 8.4620 Accuracy 0.0424
Epoch 1 Batch 250 Loss 8.2929 Accuracy 0.0474
Epoch 1 Batch 300 Loss 8.1045 Accuracy 0.0522
Epoch 1 Batch 350 Loss 7.9074 Accuracy 0.0579
Epoch 1 Batch 400 Loss 7.7174 Accuracy 0.0657
Epoch 1 Batch 450 Loss 7.5503 Accuracy 0.0728
Epoch 1 Batch 500 Loss 7.4023 Accuracy 0.0794
Epoch 1 Batch 550 Loss 7.2690 Accuracy 0.0864
Epoch 1 Batch 600 Loss 7.1429 Accuracy 0.0939
Epoch 1 Batch 650 Loss 7.0231 Accuracy 0.1012
Epoch 1 Batch 700 Loss 6.9124 Accuracy 0.1077
Epoch 1 Batch 750 Loss 6.8118 Accuracy 0.1140
Epoch 1 Batch 800 Loss 6.7182 Accuracy 0.1195
Epoch 1 Loss 6.7023 Accuracy 0.1205
Time taken for 1 epoch: 67.89 secs

Epoch 2 Batch 0 Loss 5.1591 Accuracy 0.2270
Epoch 2 Batch 50 Loss 5.2163 Accuracy 0.2126
Epoch 2 Batch 100 Loss 5.1807 Accuracy 0.2167
Epoch 2 Batch 150 Loss 5.1651 Accuracy 0.2190
Epoch 2 Batch 200 Loss 5.1359 Accuracy 0.2227
Epoch 2 Batch 250 Loss 5.1069 Accuracy 0.2253
Epoch 2 Batch 300 Loss 5.0841 Accuracy 0.2275
Epoch 2 Batch 350 Loss 5.0592 Accuracy 0.2301
Epoch 2 Batch 400 Loss 5.0360 Accuracy 0.2325
Epoch 2 Batch 450 Loss 5.0155 Accuracy 0.2346
Epoch 2 Batch 500 Loss 4.9966 Accuracy 0.2364
Epoch 2 Batch 550 Loss 4.9763 Accuracy 0.2382
Epoch 2 Batch 600 Loss 4.9571 Accuracy 0.2400
Epoch 2 Batch 650 Loss 4.9392 Accuracy 0.2417
Epoch 2 Batch 700 Loss 4.9229 Accuracy 0.2431
Epoch 2 Batch 750 Loss 4.9063 Accuracy 0.2445
Epoch 2 Batch 800 Loss 4.8908 Accuracy 0.2458
Epoch 2 Loss 4.8871 Accuracy 0.2462
Time taken for 1 epoch: 51.84 secs

Epoch 3 Batch 0 Loss 4.7496 Accuracy 0.2528
Epoch 3 Batch 50 Loss 4.5931 Accuracy 0.2724
Epoch 3 Batch 100 Loss 4.5815 Accuracy 0.2730
Epoch 3 Batch 150 Loss 4.5607 Accuracy 0.2747
Epoch 3 Batch 200 Loss 4.5467 Accuracy 0.2760
Epoch 3 Batch 250 Loss 4.5444 Accuracy 0.2764
Epoch 3 Batch 300 Loss 4.5312 Accuracy 0.2778
Epoch 3 Batch 350 Loss 4.5188 Accuracy 0.2791
Epoch 3 Batch 400 Loss 4.5048 Accuracy 0.2808
Epoch 3 Batch 450 Loss 4.4914 Accuracy 0.2822
Epoch 3 Batch 500 Loss 4.4781 Accuracy 0.2835
Epoch 3 Batch 550 Loss 4.4637 Accuracy 0.2853
Epoch 3 Batch 600 Loss 4.4502 Accuracy 0.2869
Epoch 3 Batch 650 Loss 4.4355 Accuracy 0.2887
Epoch 3 Batch 700 Loss 4.4193 Accuracy 0.2904
Epoch 3 Batch 750 Loss 4.4046 Accuracy 0.2922
Epoch 3 Batch 800 Loss 4.3877 Accuracy 0.2943
Epoch 3 Loss 4.3835 Accuracy 0.2947
Time taken for 1 epoch: 50.60 secs

Epoch 4 Batch 0 Loss 4.1652 Accuracy 0.3333
Epoch 4 Batch 50 Loss 4.0788 Accuracy 0.3290
Epoch 4 Batch 100 Loss 4.0445 Accuracy 0.3339
Epoch 4 Batch 150 Loss 4.0198 Accuracy 0.3370
Epoch 4 Batch 200 Loss 4.0064 Accuracy 0.3390
Epoch 4 Batch 250 Loss 3.9912 Accuracy 0.3406
Epoch 4 Batch 300 Loss 3.9757 Accuracy 0.3426
Epoch 4 Batch 350 Loss 3.9589 Accuracy 0.3447
Epoch 4 Batch 400 Loss 3.9458 Accuracy 0.3464
Epoch 4 Batch 450 Loss 3.9309 Accuracy 0.3481
Epoch 4 Batch 500 Loss 3.9130 Accuracy 0.3505
Epoch 4 Batch 550 Loss 3.9002 Accuracy 0.3521
Epoch 4 Batch 600 Loss 3.8851 Accuracy 0.3542
Epoch 4 Batch 650 Loss 3.8680 Accuracy 0.3564
Epoch 4 Batch 700 Loss 3.8556 Accuracy 0.3582
Epoch 4 Batch 750 Loss 3.8442 Accuracy 0.3597
Epoch 4 Batch 800 Loss 3.8292 Accuracy 0.3618
Epoch 4 Loss 3.8263 Accuracy 0.3622
Time taken for 1 epoch: 50.77 secs

Epoch 5 Batch 0 Loss 3.6539 Accuracy 0.3988
Epoch 5 Batch 50 Loss 3.5384 Accuracy 0.3941
Epoch 5 Batch 100 Loss 3.5270 Accuracy 0.3959
Epoch 5 Batch 150 Loss 3.5129 Accuracy 0.3992
Epoch 5 Batch 200 Loss 3.4907 Accuracy 0.4023
Epoch 5 Batch 250 Loss 3.4844 Accuracy 0.4029
Epoch 5 Batch 300 Loss 3.4699 Accuracy 0.4049
Epoch 5 Batch 350 Loss 3.4615 Accuracy 0.4060
Epoch 5 Batch 400 Loss 3.4520 Accuracy 0.4073
Epoch 5 Batch 450 Loss 3.4455 Accuracy 0.4080
Epoch 5 Batch 500 Loss 3.4331 Accuracy 0.4098
Epoch 5 Batch 550 Loss 3.4182 Accuracy 0.4116
Epoch 5 Batch 600 Loss 3.4119 Accuracy 0.4124
Epoch 5 Batch 650 Loss 3.4020 Accuracy 0.4137
Epoch 5 Batch 700 Loss 3.3951 Accuracy 0.4148
Epoch 5 Batch 750 Loss 3.3880 Accuracy 0.4156
Epoch 5 Batch 800 Loss 3.3780 Accuracy 0.4171
Saving checkpoint for epoch 5 at ./checkpoints/train/ckpt-1
Epoch 5 Loss 3.3768 Accuracy 0.4172
Time taken for 1 epoch: 50.68 secs

Epoch 6 Batch 0 Loss 3.2038 Accuracy 0.4183
Epoch 6 Batch 50 Loss 3.1073 Accuracy 0.4465
Epoch 6 Batch 100 Loss 3.1056 Accuracy 0.4467
Epoch 6 Batch 150 Loss 3.0853 Accuracy 0.4500
Epoch 6 Batch 200 Loss 3.0788 Accuracy 0.4512
Epoch 6 Batch 250 Loss 3.0802 Accuracy 0.4507
Epoch 6 Batch 300 Loss 3.0750 Accuracy 0.4516
Epoch 6 Batch 350 Loss 3.0661 Accuracy 0.4528
Epoch 6 Batch 400 Loss 3.0562 Accuracy 0.4543
Epoch 6 Batch 450 Loss 3.0464 Accuracy 0.4558
Epoch 6 Batch 500 Loss 3.0342 Accuracy 0.4575
Epoch 6 Batch 550 Loss 3.0282 Accuracy 0.4584
Epoch 6 Batch 600 Loss 3.0199 Accuracy 0.4596
Epoch 6 Batch 650 Loss 3.0085 Accuracy 0.4614
Epoch 6 Batch 700 Loss 3.0005 Accuracy 0.4625
Epoch 6 Batch 750 Loss 2.9946 Accuracy 0.4634
Epoch 6 Batch 800 Loss 2.9895 Accuracy 0.4642
Epoch 6 Loss 2.9875 Accuracy 0.4645
Time taken for 1 epoch: 50.88 secs

Epoch 7 Batch 0 Loss 2.7694 Accuracy 0.4982
Epoch 7 Batch 50 Loss 2.7504 Accuracy 0.4915
Epoch 7 Batch 100 Loss 2.7268 Accuracy 0.4952
Epoch 7 Batch 150 Loss 2.7187 Accuracy 0.4960
Epoch 7 Batch 200 Loss 2.7095 Accuracy 0.4972
Epoch 7 Batch 250 Loss 2.7081 Accuracy 0.4978
Epoch 7 Batch 300 Loss 2.6940 Accuracy 0.5001
Epoch 7 Batch 350 Loss 2.6906 Accuracy 0.5008
Epoch 7 Batch 400 Loss 2.6818 Accuracy 0.5023
Epoch 7 Batch 450 Loss 2.6795 Accuracy 0.5027
Epoch 7 Batch 500 Loss 2.6768 Accuracy 0.5031
Epoch 7 Batch 550 Loss 2.6754 Accuracy 0.5034
Epoch 7 Batch 600 Loss 2.6715 Accuracy 0.5039
Epoch 7 Batch 650 Loss 2.6682 Accuracy 0.5046
Epoch 7 Batch 700 Loss 2.6643 Accuracy 0.5053
Epoch 7 Batch 750 Loss 2.6607 Accuracy 0.5060
Epoch 7 Batch 800 Loss 2.6548 Accuracy 0.5070
Epoch 7 Loss 2.6560 Accuracy 0.5069
Time taken for 1 epoch: 50.55 secs

Epoch 8 Batch 0 Loss 2.5369 Accuracy 0.5155
Epoch 8 Batch 50 Loss 2.4466 Accuracy 0.5349
Epoch 8 Batch 100 Loss 2.4255 Accuracy 0.5371
Epoch 8 Batch 150 Loss 2.4299 Accuracy 0.5357
Epoch 8 Batch 200 Loss 2.4301 Accuracy 0.5361
Epoch 8 Batch 250 Loss 2.4266 Accuracy 0.5369
Epoch 8 Batch 300 Loss 2.4291 Accuracy 0.5367
Epoch 8 Batch 350 Loss 2.4275 Accuracy 0.5372
Epoch 8 Batch 400 Loss 2.4261 Accuracy 0.5372
Epoch 8 Batch 450 Loss 2.4248 Accuracy 0.5373
Epoch 8 Batch 500 Loss 2.4253 Accuracy 0.5371
Epoch 8 Batch 550 Loss 2.4238 Accuracy 0.5374
Epoch 8 Batch 600 Loss 2.4215 Accuracy 0.5378
Epoch 8 Batch 650 Loss 2.4203 Accuracy 0.5383
Epoch 8 Batch 700 Loss 2.4182 Accuracy 0.5386
Epoch 8 Batch 750 Loss 2.4160 Accuracy 0.5389
Epoch 8 Batch 800 Loss 2.4154 Accuracy 0.5391
Epoch 8 Loss 2.4155 Accuracy 0.5391
Time taken for 1 epoch: 50.08 secs

Epoch 9 Batch 0 Loss 2.1173 Accuracy 0.5724
Epoch 9 Batch 50 Loss 2.2597 Accuracy 0.5580
Epoch 9 Batch 100 Loss 2.2266 Accuracy 0.5628
Epoch 9 Batch 150 Loss 2.2298 Accuracy 0.5626
Epoch 9 Batch 200 Loss 2.2301 Accuracy 0.5627
Epoch 9 Batch 250 Loss 2.2381 Accuracy 0.5619
Epoch 9 Batch 300 Loss 2.2342 Accuracy 0.5626
Epoch 9 Batch 350 Loss 2.2366 Accuracy 0.5623
Epoch 9 Batch 400 Loss 2.2374 Accuracy 0.5623
Epoch 9 Batch 450 Loss 2.2357 Accuracy 0.5627
Epoch 9 Batch 500 Loss 2.2348 Accuracy 0.5630
Epoch 9 Batch 550 Loss 2.2324 Accuracy 0.5635
Epoch 9 Batch 600 Loss 2.2337 Accuracy 0.5634
Epoch 9 Batch 650 Loss 2.2343 Accuracy 0.5632
Epoch 9 Batch 700 Loss 2.2358 Accuracy 0.5631
Epoch 9 Batch 750 Loss 2.2344 Accuracy 0.5634
Epoch 9 Batch 800 Loss 2.2353 Accuracy 0.5634
Epoch 9 Loss 2.2343 Accuracy 0.5636
Time taken for 1 epoch: 48.99 secs

Epoch 10 Batch 0 Loss 1.9922 Accuracy 0.5977
Epoch 10 Batch 50 Loss 2.0634 Accuracy 0.5875
Epoch 10 Batch 100 Loss 2.0665 Accuracy 0.5864
Epoch 10 Batch 150 Loss 2.0772 Accuracy 0.5852
Epoch 10 Batch 200 Loss 2.0844 Accuracy 0.5847
Epoch 10 Batch 250 Loss 2.0829 Accuracy 0.5845
Epoch 10 Batch 300 Loss 2.0851 Accuracy 0.5841
Epoch 10 Batch 350 Loss 2.0884 Accuracy 0.5838
Epoch 10 Batch 400 Loss 2.0882 Accuracy 0.5838
Epoch 10 Batch 450 Loss 2.0893 Accuracy 0.5835
Epoch 10 Batch 500 Loss 2.0865 Accuracy 0.5839
Epoch 10 Batch 550 Loss 2.0883 Accuracy 0.5839
Epoch 10 Batch 600 Loss 2.0881 Accuracy 0.5840
Epoch 10 Batch 650 Loss 2.0886 Accuracy 0.5839
Epoch 10 Batch 700 Loss 2.0903 Accuracy 0.5839
Epoch 10 Batch 750 Loss 2.0901 Accuracy 0.5839
Epoch 10 Batch 800 Loss 2.0936 Accuracy 0.5836
Saving checkpoint for epoch 10 at ./checkpoints/train/ckpt-2
Epoch 10 Loss 2.0936 Accuracy 0.5836
Time taken for 1 epoch: 49.90 secs

Epoch 11 Batch 0 Loss 1.9743 Accuracy 0.6003
Epoch 11 Batch 50 Loss 1.9758 Accuracy 0.5977
Epoch 11 Batch 100 Loss 1.9516 Accuracy 0.6016
Epoch 11 Batch 150 Loss 1.9523 Accuracy 0.6028
Epoch 11 Batch 200 Loss 1.9650 Accuracy 0.6004
Epoch 11 Batch 250 Loss 1.9619 Accuracy 0.6013
Epoch 11 Batch 300 Loss 1.9623 Accuracy 0.6012
Epoch 11 Batch 350 Loss 1.9674 Accuracy 0.6003
Epoch 11 Batch 400 Loss 1.9678 Accuracy 0.6003
Epoch 11 Batch 450 Loss 1.9687 Accuracy 0.6004
Epoch 11 Batch 500 Loss 1.9708 Accuracy 0.6001
Epoch 11 Batch 550 Loss 1.9738 Accuracy 0.5997
Epoch 11 Batch 600 Loss 1.9769 Accuracy 0.5994
Epoch 11 Batch 650 Loss 1.9754 Accuracy 0.5997
Epoch 11 Batch 700 Loss 1.9760 Accuracy 0.5998
Epoch 11 Batch 750 Loss 1.9789 Accuracy 0.5994
Epoch 11 Batch 800 Loss 1.9801 Accuracy 0.5994
Epoch 11 Loss 1.9801 Accuracy 0.5993
Time taken for 1 epoch: 50.51 secs

Epoch 12 Batch 0 Loss 2.0371 Accuracy 0.5814
Epoch 12 Batch 50 Loss 1.8457 Accuracy 0.6189
Epoch 12 Batch 100 Loss 1.8479 Accuracy 0.6178
Epoch 12 Batch 150 Loss 1.8477 Accuracy 0.6182
Epoch 12 Batch 200 Loss 1.8504 Accuracy 0.6180
Epoch 12 Batch 250 Loss 1.8545 Accuracy 0.6172
Epoch 12 Batch 300 Loss 1.8621 Accuracy 0.6159
Epoch 12 Batch 350 Loss 1.8635 Accuracy 0.6157
Epoch 12 Batch 400 Loss 1.8686 Accuracy 0.6153
Epoch 12 Batch 450 Loss 1.8712 Accuracy 0.6151
Epoch 12 Batch 500 Loss 1.8746 Accuracy 0.6145
Epoch 12 Batch 550 Loss 1.8756 Accuracy 0.6143
Epoch 12 Batch 600 Loss 1.8764 Accuracy 0.6143
Epoch 12 Batch 650 Loss 1.8796 Accuracy 0.6136
Epoch 12 Batch 700 Loss 1.8810 Accuracy 0.6135
Epoch 12 Batch 750 Loss 1.8836 Accuracy 0.6134
Epoch 12 Batch 800 Loss 1.8847 Accuracy 0.6133
Epoch 12 Loss 1.8841 Accuracy 0.6134
Time taken for 1 epoch: 50.77 secs

Epoch 13 Batch 0 Loss 1.7012 Accuracy 0.6425
Epoch 13 Batch 50 Loss 1.7877 Accuracy 0.6252
Epoch 13 Batch 100 Loss 1.7753 Accuracy 0.6284
Epoch 13 Batch 150 Loss 1.7830 Accuracy 0.6275
Epoch 13 Batch 200 Loss 1.7758 Accuracy 0.6285
Epoch 13 Batch 250 Loss 1.7815 Accuracy 0.6274
Epoch 13 Batch 300 Loss 1.7873 Accuracy 0.6266
Epoch 13 Batch 350 Loss 1.7875 Accuracy 0.6266
Epoch 13 Batch 400 Loss 1.7876 Accuracy 0.6268
Epoch 13 Batch 450 Loss 1.7885 Accuracy 0.6266
Epoch 13 Batch 500 Loss 1.7904 Accuracy 0.6264
Epoch 13 Batch 550 Loss 1.7896 Accuracy 0.6266
Epoch 13 Batch 600 Loss 1.7909 Accuracy 0.6266
Epoch 13 Batch 650 Loss 1.7947 Accuracy 0.6260
Epoch 13 Batch 700 Loss 1.7988 Accuracy 0.6255
Epoch 13 Batch 750 Loss 1.8013 Accuracy 0.6254
Epoch 13 Batch 800 Loss 1.8027 Accuracy 0.6253
Epoch 13 Loss 1.8044 Accuracy 0.6251
Time taken for 1 epoch: 50.61 secs

Epoch 14 Batch 0 Loss 1.5408 Accuracy 0.6717
Epoch 14 Batch 50 Loss 1.6916 Accuracy 0.6429
Epoch 14 Batch 100 Loss 1.6907 Accuracy 0.6429
Epoch 14 Batch 150 Loss 1.7023 Accuracy 0.6407
Epoch 14 Batch 200 Loss 1.7057 Accuracy 0.6400
Epoch 14 Batch 250 Loss 1.7103 Accuracy 0.6390
Epoch 14 Batch 300 Loss 1.7094 Accuracy 0.6392
Epoch 14 Batch 350 Loss 1.7124 Accuracy 0.6389
Epoch 14 Batch 400 Loss 1.7136 Accuracy 0.6387
Epoch 14 Batch 450 Loss 1.7161 Accuracy 0.6383
Epoch 14 Batch 500 Loss 1.7169 Accuracy 0.6381
Epoch 14 Batch 550 Loss 1.7184 Accuracy 0.6380
Epoch 14 Batch 600 Loss 1.7204 Accuracy 0.6376
Epoch 14 Batch 650 Loss 1.7242 Accuracy 0.6370
Epoch 14 Batch 700 Loss 1.7270 Accuracy 0.6365
Epoch 14 Batch 750 Loss 1.7311 Accuracy 0.6359
Epoch 14 Batch 800 Loss 1.7329 Accuracy 0.6357
Epoch 14 Loss 1.7332 Accuracy 0.6357
Time taken for 1 epoch: 49.79 secs

Epoch 15 Batch 0 Loss 1.6071 Accuracy 0.6600
Epoch 15 Batch 50 Loss 1.6075 Accuracy 0.6563
Epoch 15 Batch 100 Loss 1.6176 Accuracy 0.6543
Epoch 15 Batch 150 Loss 1.6315 Accuracy 0.6522
Epoch 15 Batch 200 Loss 1.6394 Accuracy 0.6505
Epoch 15 Batch 250 Loss 1.6374 Accuracy 0.6508
Epoch 15 Batch 300 Loss 1.6396 Accuracy 0.6503
Epoch 15 Batch 350 Loss 1.6453 Accuracy 0.6494
Epoch 15 Batch 400 Loss 1.6499 Accuracy 0.6485
Epoch 15 Batch 450 Loss 1.6535 Accuracy 0.6480
Epoch 15 Batch 500 Loss 1.6549 Accuracy 0.6479
Epoch 15 Batch 550 Loss 1.6592 Accuracy 0.6470
Epoch 15 Batch 600 Loss 1.6633 Accuracy 0.6464
Epoch 15 Batch 650 Loss 1.6653 Accuracy 0.6462
Epoch 15 Batch 700 Loss 1.6672 Accuracy 0.6460
Epoch 15 Batch 750 Loss 1.6705 Accuracy 0.6455
Epoch 15 Batch 800 Loss 1.6713 Accuracy 0.6454
Saving checkpoint for epoch 15 at ./checkpoints/train/ckpt-3
Epoch 15 Loss 1.6711 Accuracy 0.6455
Time taken for 1 epoch: 49.58 secs

Epoch 16 Batch 0 Loss 1.5173 Accuracy 0.6608
Epoch 16 Batch 50 Loss 1.5766 Accuracy 0.6589
Epoch 16 Batch 100 Loss 1.5695 Accuracy 0.6607
Epoch 16 Batch 150 Loss 1.5790 Accuracy 0.6588
Epoch 16 Batch 200 Loss 1.5813 Accuracy 0.6589
Epoch 16 Batch 250 Loss 1.5881 Accuracy 0.6578
Epoch 16 Batch 300 Loss 1.5924 Accuracy 0.6570
Epoch 16 Batch 350 Loss 1.5967 Accuracy 0.6565
Epoch 16 Batch 400 Loss 1.5992 Accuracy 0.6560
Epoch 16 Batch 450 Loss 1.6021 Accuracy 0.6554
Epoch 16 Batch 500 Loss 1.6016 Accuracy 0.6557
Epoch 16 Batch 550 Loss 1.6044 Accuracy 0.6554
Epoch 16 Batch 600 Loss 1.6072 Accuracy 0.6548
Epoch 16 Batch 650 Loss 1.6100 Accuracy 0.6544
Epoch 16 Batch 700 Loss 1.6125 Accuracy 0.6540
Epoch 16 Batch 750 Loss 1.6164 Accuracy 0.6536
Epoch 16 Batch 800 Loss 1.6184 Accuracy 0.6534
Epoch 16 Loss 1.6194 Accuracy 0.6532
Time taken for 1 epoch: 49.55 secs

Epoch 17 Batch 0 Loss 1.4743 Accuracy 0.6809
Epoch 17 Batch 50 Loss 1.5117 Accuracy 0.6702
Epoch 17 Batch 100 Loss 1.5143 Accuracy 0.6695
Epoch 17 Batch 150 Loss 1.5157 Accuracy 0.6699
Epoch 17 Batch 200 Loss 1.5319 Accuracy 0.6668
Epoch 17 Batch 250 Loss 1.5337 Accuracy 0.6664
Epoch 17 Batch 300 Loss 1.5353 Accuracy 0.6663
Epoch 17 Batch 350 Loss 1.5389 Accuracy 0.6658
Epoch 17 Batch 400 Loss 1.5407 Accuracy 0.6655
Epoch 17 Batch 450 Loss 1.5454 Accuracy 0.6646
Epoch 17 Batch 500 Loss 1.5465 Accuracy 0.6644
Epoch 17 Batch 550 Loss 1.5507 Accuracy 0.6638
Epoch 17 Batch 600 Loss 1.5543 Accuracy 0.6634
Epoch 17 Batch 650 Loss 1.5579 Accuracy 0.6629
Epoch 17 Batch 700 Loss 1.5602 Accuracy 0.6625
Epoch 17 Batch 750 Loss 1.5646 Accuracy 0.6619
Epoch 17 Batch 800 Loss 1.5667 Accuracy 0.6614
Epoch 17 Loss 1.5678 Accuracy 0.6613
Time taken for 1 epoch: 49.26 secs

Epoch 18 Batch 0 Loss 1.5091 Accuracy 0.6582
Epoch 18 Batch 50 Loss 1.4854 Accuracy 0.6739
Epoch 18 Batch 100 Loss 1.4776 Accuracy 0.6749
Epoch 18 Batch 150 Loss 1.4790 Accuracy 0.6746
Epoch 18 Batch 200 Loss 1.4861 Accuracy 0.6735
Epoch 18 Batch 250 Loss 1.4875 Accuracy 0.6736
Epoch 18 Batch 300 Loss 1.4922 Accuracy 0.6730
Epoch 18 Batch 350 Loss 1.4953 Accuracy 0.6725
Epoch 18 Batch 400 Loss 1.5018 Accuracy 0.6714
Epoch 18 Batch 450 Loss 1.5055 Accuracy 0.6710
Epoch 18 Batch 500 Loss 1.5063 Accuracy 0.6710
Epoch 18 Batch 550 Loss 1.5091 Accuracy 0.6705
Epoch 18 Batch 600 Loss 1.5107 Accuracy 0.6701
Epoch 18 Batch 650 Loss 1.5142 Accuracy 0.6695
Epoch 18 Batch 700 Loss 1.5175 Accuracy 0.6690
Epoch 18 Batch 750 Loss 1.5198 Accuracy 0.6687
Epoch 18 Batch 800 Loss 1.5233 Accuracy 0.6683
Epoch 18 Loss 1.5240 Accuracy 0.6682
Time taken for 1 epoch: 49.24 secs

Epoch 19 Batch 0 Loss 1.3142 Accuracy 0.6876
Epoch 19 Batch 50 Loss 1.4266 Accuracy 0.6843
Epoch 19 Batch 100 Loss 1.4270 Accuracy 0.6841
Epoch 19 Batch 150 Loss 1.4367 Accuracy 0.6822
Epoch 19 Batch 200 Loss 1.4445 Accuracy 0.6810
Epoch 19 Batch 250 Loss 1.4516 Accuracy 0.6796
Epoch 19 Batch 300 Loss 1.4517 Accuracy 0.6799
Epoch 19 Batch 350 Loss 1.4592 Accuracy 0.6786
Epoch 19 Batch 400 Loss 1.4637 Accuracy 0.6777
Epoch 19 Batch 450 Loss 1.4640 Accuracy 0.6776
Epoch 19 Batch 500 Loss 1.4669 Accuracy 0.6773
Epoch 19 Batch 550 Loss 1.4672 Accuracy 0.6771
Epoch 19 Batch 600 Loss 1.4714 Accuracy 0.6764
Epoch 19 Batch 650 Loss 1.4766 Accuracy 0.6755
Epoch 19 Batch 700 Loss 1.4781 Accuracy 0.6752
Epoch 19 Batch 750 Loss 1.4817 Accuracy 0.6746
Epoch 19 Batch 800 Loss 1.4834 Accuracy 0.6745
Epoch 19 Loss 1.4833 Accuracy 0.6745
Time taken for 1 epoch: 48.96 secs

Epoch 20 Batch 0 Loss 1.4669 Accuracy 0.6734
Epoch 20 Batch 50 Loss 1.3892 Accuracy 0.6880
Epoch 20 Batch 100 Loss 1.3869 Accuracy 0.6889
Epoch 20 Batch 150 Loss 1.3953 Accuracy 0.6876
Epoch 20 Batch 200 Loss 1.3963 Accuracy 0.6882
Epoch 20 Batch 250 Loss 1.4018 Accuracy 0.6877
Epoch 20 Batch 300 Loss 1.4066 Accuracy 0.6868
Epoch 20 Batch 350 Loss 1.4136 Accuracy 0.6857
Epoch 20 Batch 400 Loss 1.4175 Accuracy 0.6851
Epoch 20 Batch 450 Loss 1.4217 Accuracy 0.6842
Epoch 20 Batch 500 Loss 1.4278 Accuracy 0.6833
Epoch 20 Batch 550 Loss 1.4296 Accuracy 0.6832
Epoch 20 Batch 600 Loss 1.4323 Accuracy 0.6827
Epoch 20 Batch 650 Loss 1.4341 Accuracy 0.6825
Epoch 20 Batch 700 Loss 1.4372 Accuracy 0.6820
Epoch 20 Batch 750 Loss 1.4408 Accuracy 0.6815
Epoch 20 Batch 800 Loss 1.4451 Accuracy 0.6807
Saving checkpoint for epoch 20 at ./checkpoints/train/ckpt-4
Epoch 20 Loss 1.4464 Accuracy 0.6806
Time taken for 1 epoch: 49.74 secs

Evaluate

The following steps are used for evaluation:

  • Encode the input sentence using the Portuguese tokenizer (tokenizers.pt). This is the encoder input.
  • The decoder input is initialized to the [START] token.
  • Calculate the padding masks and the look ahead masks.
  • The decoder then outputs the predictions by looking at the encoder output and its own output (self-attention).
  • The model makes predictions of the next word for each word in the output. Most of these are redundant. Use the predictions from the last word.
  • Concatenate the predicted word to the decoder input and pass it to the decoder.
  • In this approach, the decoder predicts the next word based on the previous words it predicted.
def evaluate(sentence, max_length=40):
  # inp sentence is portuguese, hence adding the start and end token
  sentence = tf.convert_to_tensor([sentence])
  sentence = tokenizers.pt.tokenize(sentence).to_tensor()

  encoder_input = sentence

  # as the target is english, the first word to the transformer should be the
  # english start token.
  start, end = tokenizers.en.tokenize([''])[0]
  output = tf.convert_to_tensor([start])
  output = tf.expand_dims(output, 0)

  for i in range(max_length):
    enc_padding_mask, combined_mask, dec_padding_mask = create_masks(
        encoder_input, output)

    # predictions.shape == (batch_size, seq_len, vocab_size)
    predictions, attention_weights = transformer(encoder_input,
                                                 output,
                                                 False,
                                                 enc_padding_mask,
                                                 combined_mask,
                                                 dec_padding_mask)

    # select the last word from the seq_len dimension
    predictions = predictions[:, -1:, :]  # (batch_size, 1, vocab_size)

    predicted_id = tf.argmax(predictions, axis=-1)

    # concatentate the predicted_id to the output which is given to the decoder
    # as its input.
    output = tf.concat([output, predicted_id], axis=-1)

    # return the result if the predicted_id is equal to the end token
    if predicted_id == end:
      break

  # output.shape (1, tokens)
  text = tokenizers.en.detokenize(output)[0]  # shape: ()

  tokens = tokenizers.en.lookup(output)[0]

  return text, tokens, attention_weights

def print_translation(sentence, tokens, ground_truth):
  print(f'{"Input:":15s}: {sentence}')
  print(f'{"Prediction":15s}: {tokens.numpy().decode("utf-8")}')
  print(f'{"Ground truth":15s}: {ground_truth}')

sentence = "este é um problema que temos que resolver."
ground_truth = "this is a problem we have to solve ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : este é um problema que temos que resolver.
Prediction     : this is a problem that we have to solve .
Ground truth   : this is a problem we have to solve .

sentence = "os meus vizinhos ouviram sobre esta ideia."
ground_truth = "and my neighboring homes heard about this idea ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : os meus vizinhos ouviram sobre esta ideia.
Prediction     : my neighbors heard about this idea .
Ground truth   : and my neighboring homes heard about this idea .

sentence = "vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram."
ground_truth = "so i \'ll just share with you some stories very quickly of some magical things that have happened ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram.
Prediction     : so i ' m going to share with you some very magical things that have happened .
Ground truth   : so i 'll just share with you some stories very quickly of some magical things that have happened .

You can pass different layers and attention blocks of the decoder to the plot parameter.

Attention plots

The evaluate function also returns a dictionary of attention maps you can use to visualize the internal working of the model:

sentence = "este é o primeiro livro que eu fiz."
ground_truth = "this is the first book i've ever done."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : este é o primeiro livro que eu fiz.
Prediction     : this is the first book i did .
Ground truth   : this is the first book i've ever done.

def plot_attention_head(in_tokens, translated_tokens, attention):
  # The plot is of the attention when a token was generated.
  # The model didn't generate `<START>` in the output. Skip it.
  translated_tokens = translated_tokens[1:]

  ax = plt.gca()
  ax.matshow(attention)
  ax.set_xticks(range(len(in_tokens)))
  ax.set_yticks(range(len(translated_tokens)))

  labels = [label.decode('utf-8') for label in in_tokens.numpy()]
  ax.set_xticklabels(
      labels, rotation=90)

  labels = [label.decode('utf-8') for label in translated_tokens.numpy()]
  ax.set_yticklabels(labels)

head = 0
# shape: (batch=1, num_heads, seq_len_q, seq_len_k)
attention_heads = tf.squeeze(
  attention_weights['decoder_layer4_block2'], 0)
attention = attention_heads[head]
attention.shape

TensorShape([9, 11])

in_tokens = tf.convert_to_tensor([sentence])
in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
in_tokens = tokenizers.pt.lookup(in_tokens)[0]
in_tokens

<tf.Tensor: shape=(11,), dtype=string, numpy=
array([b'[START]', b'este', b'e', b'o', b'primeiro', b'livro', b'que',
       b'eu', b'fiz', b'.', b'[END]'], dtype=object)>

translated_tokens

<tf.Tensor: shape=(10,), dtype=string, numpy=
array([b'[START]', b'this', b'is', b'the', b'first', b'book', b'i',
       b'did', b'.', b'[END]'], dtype=object)>

plot_attention_head(in_tokens, translated_tokens, attention)

png

def plot_attention_weights(sentence, translated_tokens, attention_heads):
  in_tokens = tf.convert_to_tensor([sentence])
  in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
  in_tokens = tokenizers.pt.lookup(in_tokens)[0]
  in_tokens

  fig = plt.figure(figsize=(16, 8))

  for h, head in enumerate(attention_heads):
    ax = fig.add_subplot(2, 4, h+1)

    plot_attention_head(in_tokens, translated_tokens, head)

    ax.set_xlabel(f'Head {h+1}')

  plt.tight_layout()
  plt.show()

plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])

png

The model does okay on unfamiliar words. Neither "triceratops" or "encyclopedia" are in the input dataset and the model almost learns to transliterate them, even without a shared vocabulary:

sentence = "Eu li sobre triceratops na enciclopédia."
ground_truth = "I read about triceratops in the encyclopedia."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])

Input:         : Eu li sobre triceratops na enciclopédia.
Prediction     : i read about trifters in egypt .
Ground truth   : I read about triceratops in the encyclopedia.

png

Summary

In this tutorial, you learned about positional encoding, multi-head attention, the importance of masking and how to create a transformer.

Try using a different dataset to train the transformer. You can also create the base transformer or transformer XL by changing the hyperparameters above. You can also use the layers defined here to create BERT and train state of the art models. Furthermore, you can implement beam search to get better predictions.